Add sort :per_call option to tprof

[sabiwara]

A small follow-up for #13605, I didn't include it in the MVP to keep the PR smaller.

One of the niceties of tprof is the ability to sort per "per call measurement", perhaps we can expose the option?
We might consider some other options as well, but sorting by percentage seems redundant since it should be proportional?

$ mix profile.tprof -e "Enum.each(1..5, &String.Chars.Integer.to_string/1)" --sort per_call
Warmup...


Profile results of #PID<0.106.0>
#                                               CALLS      % TIME µS/CALL
Total                                              20 100.00    6    0.30
String.Chars.Integer.to_string/1                    5   0.00    0    0.00
Enum.reduce_range/5                                 3   0.00    0    0.00
:erlang.integer_to_binary/1                         5  16.67    1    0.20
anonymous fn/3 in Enum.each/2                       5  33.33    2    0.40
anonymous fn/0 in :elixir_compiler_1.__FILE__/1     1  16.67    1    1.00
Enum.each/2                                         1  33.33    2    2.00

Profile done over 6 matching functions

mix profile.tprof -e "Enum.each(1..5, &String.Chars.Integer.to_string/1)" --type memory --sort per_call
Warmup...


Profile results of #PID<0.106.0>
#                           CALLS      % WORDS PER CALL
Total                           6 100.00    19     3.17
:erlang.integer_to_binary/1     5  78.95    15     3.00
Enum.each/2                     1  21.05     4     4.00

Profile done over 2 matching functions

[sabiwara]

Note: we could add a when type != :calls here because it doesn't really make sense and would just return arbitrary order, but since tprof allows it I thought there is no need to be extra defensive.

[sabiwara]

Note: prepending \n here because otherwise it might match due to anonymous fn/3 in Enum.each/2

[josevalim]

Please backport to v1.17 after merging!

[sabiwara]

Backported ✅